1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Definitions for the TCP module. 7 * 8 * Version: @(#)tcp.h 1.0.5 05/23/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License 15 * as published by the Free Software Foundation; either version 16 * 2 of the License, or (at your option) any later version. 17 */ 18 #ifndef _TCP_H 19 #define _TCP_H 20 21 #define FASTRETRANS_DEBUG 1 22 23 #include <linux/list.h> 24 #include <linux/tcp.h> 25 #include <linux/bug.h> 26 #include <linux/slab.h> 27 #include <linux/cache.h> 28 #include <linux/percpu.h> 29 #include <linux/skbuff.h> 30 #include <linux/dmaengine.h> 31 #include <linux/crypto.h> 32 #include <linux/cryptohash.h> 33 #include <linux/kref.h> 34 35 #include <net/inet_connection_sock.h> 36 #include <net/inet_timewait_sock.h> 37 #include <net/inet_hashtables.h> 38 #include <net/checksum.h> 39 #include <net/request_sock.h> 40 #include <net/sock.h> 41 #include <net/snmp.h> 42 #include <net/ip.h> 43 #include <net/tcp_states.h> 44 #include <net/inet_ecn.h> 45 #include <net/dst.h> 46 47 #include <linux/seq_file.h> 48 #include <linux/memcontrol.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 extern struct percpu_counter tcp_orphan_count; 53 extern void tcp_time_wait(struct sock *sk, int state, int timeo); 54 55 #define MAX_TCP_HEADER (128 + MAX_HEADER) 56 #define MAX_TCP_OPTION_SPACE 40 57 58 /* 59 * Never offer a window over 32767 without using window scaling. Some 60 * poor stacks do signed 16bit maths! 61 */ 62 #define MAX_TCP_WINDOW 32767U 63 64 /* Offer an initial receive window of 10 mss. */ 65 #define TCP_DEFAULT_INIT_RCVWND 10 66 67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 68 #define TCP_MIN_MSS 88U 69 70 /* The least MTU to use for probing */ 71 #define TCP_BASE_MSS 512 72 73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 74 #define TCP_FASTRETRANS_THRESH 3 75 76 /* Maximal reordering. */ 77 #define TCP_MAX_REORDERING 127 78 79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 80 #define TCP_MAX_QUICKACKS 16U 81 82 /* urg_data states */ 83 #define TCP_URG_VALID 0x0100 84 #define TCP_URG_NOTYET 0x0200 85 #define TCP_URG_READ 0x0400 86 87 #define TCP_RETR1 3 /* 88 * This is how many retries it does before it 89 * tries to figure out if the gateway is 90 * down. Minimal RFC value is 3; it corresponds 91 * to ~3sec-8min depending on RTO. 92 */ 93 94 #define TCP_RETR2 15 /* 95 * This should take at least 96 * 90 minutes to time out. 97 * RFC1122 says that the limit is 100 sec. 98 * 15 is ~13-30min depending on RTO. 99 */ 100 101 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 102 * connection: ~180sec is RFC minimum */ 103 104 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 105 * connection: ~180sec is RFC minimum */ 106 107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 108 * state, about 60 seconds */ 109 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 110 /* BSD style FIN_WAIT2 deadlock breaker. 111 * It used to be 3min, new value is 60sec, 112 * to combine FIN-WAIT-2 timeout with 113 * TIME-WAIT timer. 114 */ 115 116 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 117 #if HZ >= 100 118 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 119 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 120 #else 121 #define TCP_DELACK_MIN 4U 122 #define TCP_ATO_MIN 4U 123 #endif 124 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 125 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 126 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC2988bis initial RTO value */ 127 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 128 * used as a fallback RTO for the 129 * initial data transmission if no 130 * valid RTT sample has been acquired, 131 * most likely due to retrans in 3WHS. 132 */ 133 134 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 135 * for local resources. 136 */ 137 138 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 139 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 140 #define TCP_KEEPALIVE_INTVL (75*HZ) 141 142 #define MAX_TCP_KEEPIDLE 32767 143 #define MAX_TCP_KEEPINTVL 32767 144 #define MAX_TCP_KEEPCNT 127 145 #define MAX_TCP_SYNCNT 127 146 147 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 148 149 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 150 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 151 * after this time. It should be equal 152 * (or greater than) TCP_TIMEWAIT_LEN 153 * to provide reliability equal to one 154 * provided by timewait state. 155 */ 156 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 157 * timestamps. It must be less than 158 * minimal timewait lifetime. 159 */ 160 /* 161 * TCP option 162 */ 163 164 #define TCPOPT_NOP 1 /* Padding */ 165 #define TCPOPT_EOL 0 /* End of options */ 166 #define TCPOPT_MSS 2 /* Segment size negotiating */ 167 #define TCPOPT_WINDOW 3 /* Window scaling */ 168 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 169 #define TCPOPT_SACK 5 /* SACK Block */ 170 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 171 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 172 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */ 173 174 /* 175 * TCP option lengths 176 */ 177 178 #define TCPOLEN_MSS 4 179 #define TCPOLEN_WINDOW 3 180 #define TCPOLEN_SACK_PERM 2 181 #define TCPOLEN_TIMESTAMP 10 182 #define TCPOLEN_MD5SIG 18 183 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */ 184 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */ 185 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN) 186 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX) 187 188 /* But this is what stacks really send out. */ 189 #define TCPOLEN_TSTAMP_ALIGNED 12 190 #define TCPOLEN_WSCALE_ALIGNED 4 191 #define TCPOLEN_SACKPERM_ALIGNED 4 192 #define TCPOLEN_SACK_BASE 2 193 #define TCPOLEN_SACK_BASE_ALIGNED 4 194 #define TCPOLEN_SACK_PERBLOCK 8 195 #define TCPOLEN_MD5SIG_ALIGNED 20 196 #define TCPOLEN_MSS_ALIGNED 4 197 198 /* Flags in tp->nonagle */ 199 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 200 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 201 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 202 203 /* TCP thin-stream limits */ 204 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 205 206 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 207 #define TCP_INIT_CWND 10 208 209 extern struct inet_timewait_death_row tcp_death_row; 210 211 /* sysctl variables for tcp */ 212 extern int sysctl_tcp_timestamps; 213 extern int sysctl_tcp_window_scaling; 214 extern int sysctl_tcp_sack; 215 extern int sysctl_tcp_fin_timeout; 216 extern int sysctl_tcp_keepalive_time; 217 extern int sysctl_tcp_keepalive_probes; 218 extern int sysctl_tcp_keepalive_intvl; 219 extern int sysctl_tcp_syn_retries; 220 extern int sysctl_tcp_synack_retries; 221 extern int sysctl_tcp_retries1; 222 extern int sysctl_tcp_retries2; 223 extern int sysctl_tcp_orphan_retries; 224 extern int sysctl_tcp_syncookies; 225 extern int sysctl_tcp_retrans_collapse; 226 extern int sysctl_tcp_stdurg; 227 extern int sysctl_tcp_rfc1337; 228 extern int sysctl_tcp_abort_on_overflow; 229 extern int sysctl_tcp_max_orphans; 230 extern int sysctl_tcp_fack; 231 extern int sysctl_tcp_reordering; 232 extern int sysctl_tcp_ecn; 233 extern int sysctl_tcp_dsack; 234 extern int sysctl_tcp_wmem[3]; 235 extern int sysctl_tcp_rmem[3]; 236 extern int sysctl_tcp_app_win; 237 extern int sysctl_tcp_adv_win_scale; 238 extern int sysctl_tcp_tw_reuse; 239 extern int sysctl_tcp_frto; 240 extern int sysctl_tcp_frto_response; 241 extern int sysctl_tcp_low_latency; 242 extern int sysctl_tcp_dma_copybreak; 243 extern int sysctl_tcp_nometrics_save; 244 extern int sysctl_tcp_moderate_rcvbuf; 245 extern int sysctl_tcp_tso_win_divisor; 246 extern int sysctl_tcp_abc; 247 extern int sysctl_tcp_mtu_probing; 248 extern int sysctl_tcp_base_mss; 249 extern int sysctl_tcp_workaround_signed_windows; 250 extern int sysctl_tcp_slow_start_after_idle; 251 extern int sysctl_tcp_max_ssthresh; 252 extern int sysctl_tcp_cookie_size; 253 extern int sysctl_tcp_thin_linear_timeouts; 254 extern int sysctl_tcp_thin_dupack; 255 256 extern atomic_long_t tcp_memory_allocated; 257 extern struct percpu_counter tcp_sockets_allocated; 258 extern int tcp_memory_pressure; 259 260 /* 261 * The next routines deal with comparing 32 bit unsigned ints 262 * and worry about wraparound (automatic with unsigned arithmetic). 263 */ 264 265 static inline int before(__u32 seq1, __u32 seq2) 266 { 267 return (__s32)(seq1-seq2) < 0; 268 } 269 #define after(seq2, seq1) before(seq1, seq2) 270 271 /* is s2<=s1<=s3 ? */ 272 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 273 { 274 return seq3 - seq2 >= seq1 - seq2; 275 } 276 277 static inline bool tcp_out_of_memory(struct sock *sk) 278 { 279 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 280 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 281 return true; 282 return false; 283 } 284 285 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 286 { 287 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 288 int orphans = percpu_counter_read_positive(ocp); 289 290 if (orphans << shift > sysctl_tcp_max_orphans) { 291 orphans = percpu_counter_sum_positive(ocp); 292 if (orphans << shift > sysctl_tcp_max_orphans) 293 return true; 294 } 295 return false; 296 } 297 298 extern bool tcp_check_oom(struct sock *sk, int shift); 299 300 /* syncookies: remember time of last synqueue overflow */ 301 static inline void tcp_synq_overflow(struct sock *sk) 302 { 303 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 304 } 305 306 /* syncookies: no recent synqueue overflow on this listening socket? */ 307 static inline int tcp_synq_no_recent_overflow(const struct sock *sk) 308 { 309 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 310 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 311 } 312 313 extern struct proto tcp_prot; 314 315 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 316 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 317 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 318 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 319 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 320 321 extern void tcp_init_mem(struct net *net); 322 323 extern void tcp_v4_err(struct sk_buff *skb, u32); 324 325 extern void tcp_shutdown (struct sock *sk, int how); 326 327 extern int tcp_v4_rcv(struct sk_buff *skb); 328 329 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it); 330 extern void *tcp_v4_tw_get_peer(struct sock *sk); 331 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 332 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 333 size_t size); 334 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 335 size_t size, int flags); 336 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 337 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 338 const struct tcphdr *th, unsigned int len); 339 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 340 const struct tcphdr *th, unsigned int len); 341 extern void tcp_rcv_space_adjust(struct sock *sk); 342 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 343 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 344 extern void tcp_twsk_destructor(struct sock *sk); 345 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 346 struct pipe_inode_info *pipe, size_t len, 347 unsigned int flags); 348 349 static inline void tcp_dec_quickack_mode(struct sock *sk, 350 const unsigned int pkts) 351 { 352 struct inet_connection_sock *icsk = inet_csk(sk); 353 354 if (icsk->icsk_ack.quick) { 355 if (pkts >= icsk->icsk_ack.quick) { 356 icsk->icsk_ack.quick = 0; 357 /* Leaving quickack mode we deflate ATO. */ 358 icsk->icsk_ack.ato = TCP_ATO_MIN; 359 } else 360 icsk->icsk_ack.quick -= pkts; 361 } 362 } 363 364 #define TCP_ECN_OK 1 365 #define TCP_ECN_QUEUE_CWR 2 366 #define TCP_ECN_DEMAND_CWR 4 367 #define TCP_ECN_SEEN 8 368 369 static __inline__ void 370 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 371 { 372 if (sysctl_tcp_ecn && th->ece && th->cwr) 373 inet_rsk(req)->ecn_ok = 1; 374 } 375 376 enum tcp_tw_status { 377 TCP_TW_SUCCESS = 0, 378 TCP_TW_RST = 1, 379 TCP_TW_ACK = 2, 380 TCP_TW_SYN = 3 381 }; 382 383 384 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 385 struct sk_buff *skb, 386 const struct tcphdr *th); 387 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 388 struct request_sock *req, 389 struct request_sock **prev); 390 extern int tcp_child_process(struct sock *parent, struct sock *child, 391 struct sk_buff *skb); 392 extern int tcp_use_frto(struct sock *sk); 393 extern void tcp_enter_frto(struct sock *sk); 394 extern void tcp_enter_loss(struct sock *sk, int how); 395 extern void tcp_clear_retrans(struct tcp_sock *tp); 396 extern void tcp_update_metrics(struct sock *sk); 397 extern void tcp_close(struct sock *sk, long timeout); 398 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 399 struct poll_table_struct *wait); 400 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 401 char __user *optval, int __user *optlen); 402 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 403 char __user *optval, unsigned int optlen); 404 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 405 char __user *optval, int __user *optlen); 406 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 407 char __user *optval, unsigned int optlen); 408 extern void tcp_set_keepalive(struct sock *sk, int val); 409 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 410 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 411 size_t len, int nonblock, int flags, int *addr_len); 412 extern void tcp_parse_options(const struct sk_buff *skb, 413 struct tcp_options_received *opt_rx, const u8 **hvpp, 414 int estab); 415 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 416 417 /* 418 * TCP v4 functions exported for the inet6 API 419 */ 420 421 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 422 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 423 extern struct sock * tcp_create_openreq_child(struct sock *sk, 424 struct request_sock *req, 425 struct sk_buff *skb); 426 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 427 struct request_sock *req, 428 struct dst_entry *dst); 429 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 430 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 431 int addr_len); 432 extern int tcp_connect(struct sock *sk); 433 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 434 struct request_sock *req, 435 struct request_values *rvp); 436 extern int tcp_disconnect(struct sock *sk, int flags); 437 438 439 /* From syncookies.c */ 440 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 441 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 442 struct ip_options *opt); 443 #ifdef CONFIG_SYN_COOKIES 444 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 445 __u16 *mss); 446 #else 447 static inline __u32 cookie_v4_init_sequence(struct sock *sk, 448 struct sk_buff *skb, 449 __u16 *mss) 450 { 451 return 0; 452 } 453 #endif 454 455 extern __u32 cookie_init_timestamp(struct request_sock *req); 456 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 457 458 /* From net/ipv6/syncookies.c */ 459 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 460 #ifdef CONFIG_SYN_COOKIES 461 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 462 __u16 *mss); 463 #else 464 static inline __u32 cookie_v6_init_sequence(struct sock *sk, 465 struct sk_buff *skb, 466 __u16 *mss) 467 { 468 return 0; 469 } 470 #endif 471 /* tcp_output.c */ 472 473 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 474 int nonagle); 475 extern int tcp_may_send_now(struct sock *sk); 476 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 477 extern void tcp_retransmit_timer(struct sock *sk); 478 extern void tcp_xmit_retransmit_queue(struct sock *); 479 extern void tcp_simple_retransmit(struct sock *); 480 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 481 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 482 483 extern void tcp_send_probe0(struct sock *); 484 extern void tcp_send_partial(struct sock *); 485 extern int tcp_write_wakeup(struct sock *); 486 extern void tcp_send_fin(struct sock *sk); 487 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 488 extern int tcp_send_synack(struct sock *); 489 extern int tcp_syn_flood_action(struct sock *sk, 490 const struct sk_buff *skb, 491 const char *proto); 492 extern void tcp_push_one(struct sock *, unsigned int mss_now); 493 extern void tcp_send_ack(struct sock *sk); 494 extern void tcp_send_delayed_ack(struct sock *sk); 495 496 /* tcp_input.c */ 497 extern void tcp_cwnd_application_limited(struct sock *sk); 498 499 /* tcp_timer.c */ 500 extern void tcp_init_xmit_timers(struct sock *); 501 static inline void tcp_clear_xmit_timers(struct sock *sk) 502 { 503 inet_csk_clear_xmit_timers(sk); 504 } 505 506 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 507 extern unsigned int tcp_current_mss(struct sock *sk); 508 509 /* Bound MSS / TSO packet size with the half of the window */ 510 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 511 { 512 int cutoff; 513 514 /* When peer uses tiny windows, there is no use in packetizing 515 * to sub-MSS pieces for the sake of SWS or making sure there 516 * are enough packets in the pipe for fast recovery. 517 * 518 * On the other hand, for extremely large MSS devices, handling 519 * smaller than MSS windows in this way does make sense. 520 */ 521 if (tp->max_window >= 512) 522 cutoff = (tp->max_window >> 1); 523 else 524 cutoff = tp->max_window; 525 526 if (cutoff && pktsize > cutoff) 527 return max_t(int, cutoff, 68U - tp->tcp_header_len); 528 else 529 return pktsize; 530 } 531 532 /* tcp.c */ 533 extern void tcp_get_info(const struct sock *, struct tcp_info *); 534 535 /* Read 'sendfile()'-style from a TCP socket */ 536 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 537 unsigned int, size_t); 538 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 539 sk_read_actor_t recv_actor); 540 541 extern void tcp_initialize_rcv_mss(struct sock *sk); 542 543 extern int tcp_mtu_to_mss(const struct sock *sk, int pmtu); 544 extern int tcp_mss_to_mtu(const struct sock *sk, int mss); 545 extern void tcp_mtup_init(struct sock *sk); 546 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt); 547 548 static inline void tcp_bound_rto(const struct sock *sk) 549 { 550 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 551 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 552 } 553 554 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 555 { 556 return (tp->srtt >> 3) + tp->rttvar; 557 } 558 559 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 560 { 561 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 562 ntohl(TCP_FLAG_ACK) | 563 snd_wnd); 564 } 565 566 static inline void tcp_fast_path_on(struct tcp_sock *tp) 567 { 568 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 569 } 570 571 static inline void tcp_fast_path_check(struct sock *sk) 572 { 573 struct tcp_sock *tp = tcp_sk(sk); 574 575 if (skb_queue_empty(&tp->out_of_order_queue) && 576 tp->rcv_wnd && 577 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 578 !tp->urg_data) 579 tcp_fast_path_on(tp); 580 } 581 582 /* Compute the actual rto_min value */ 583 static inline u32 tcp_rto_min(struct sock *sk) 584 { 585 const struct dst_entry *dst = __sk_dst_get(sk); 586 u32 rto_min = TCP_RTO_MIN; 587 588 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 589 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 590 return rto_min; 591 } 592 593 /* Compute the actual receive window we are currently advertising. 594 * Rcv_nxt can be after the window if our peer push more data 595 * than the offered window. 596 */ 597 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 598 { 599 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 600 601 if (win < 0) 602 win = 0; 603 return (u32) win; 604 } 605 606 /* Choose a new window, without checks for shrinking, and without 607 * scaling applied to the result. The caller does these things 608 * if necessary. This is a "raw" window selection. 609 */ 610 extern u32 __tcp_select_window(struct sock *sk); 611 612 /* TCP timestamps are only 32-bits, this causes a slight 613 * complication on 64-bit systems since we store a snapshot 614 * of jiffies in the buffer control blocks below. We decided 615 * to use only the low 32-bits of jiffies and hide the ugly 616 * casts with the following macro. 617 */ 618 #define tcp_time_stamp ((__u32)(jiffies)) 619 620 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 621 622 #define TCPHDR_FIN 0x01 623 #define TCPHDR_SYN 0x02 624 #define TCPHDR_RST 0x04 625 #define TCPHDR_PSH 0x08 626 #define TCPHDR_ACK 0x10 627 #define TCPHDR_URG 0x20 628 #define TCPHDR_ECE 0x40 629 #define TCPHDR_CWR 0x80 630 631 /* This is what the send packet queuing engine uses to pass 632 * TCP per-packet control information to the transmission code. 633 * We also store the host-order sequence numbers in here too. 634 * This is 44 bytes if IPV6 is enabled. 635 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 636 */ 637 struct tcp_skb_cb { 638 union { 639 struct inet_skb_parm h4; 640 #if IS_ENABLED(CONFIG_IPV6) 641 struct inet6_skb_parm h6; 642 #endif 643 } header; /* For incoming frames */ 644 __u32 seq; /* Starting sequence number */ 645 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 646 __u32 when; /* used to compute rtt's */ 647 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 648 __u8 sacked; /* State flags for SACK/FACK. */ 649 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 650 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 651 #define TCPCB_LOST 0x04 /* SKB is lost */ 652 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 653 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 654 /* 1 byte hole */ 655 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 656 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 657 658 __u32 ack_seq; /* Sequence number ACK'd */ 659 }; 660 661 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 662 663 /* Due to TSO, an SKB can be composed of multiple actual 664 * packets. To keep these tracked properly, we use this. 665 */ 666 static inline int tcp_skb_pcount(const struct sk_buff *skb) 667 { 668 return skb_shinfo(skb)->gso_segs; 669 } 670 671 /* This is valid iff tcp_skb_pcount() > 1. */ 672 static inline int tcp_skb_mss(const struct sk_buff *skb) 673 { 674 return skb_shinfo(skb)->gso_size; 675 } 676 677 /* Events passed to congestion control interface */ 678 enum tcp_ca_event { 679 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 680 CA_EVENT_CWND_RESTART, /* congestion window restart */ 681 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 682 CA_EVENT_FRTO, /* fast recovery timeout */ 683 CA_EVENT_LOSS, /* loss timeout */ 684 CA_EVENT_FAST_ACK, /* in sequence ack */ 685 CA_EVENT_SLOW_ACK, /* other ack */ 686 }; 687 688 /* 689 * Interface for adding new TCP congestion control handlers 690 */ 691 #define TCP_CA_NAME_MAX 16 692 #define TCP_CA_MAX 128 693 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 694 695 #define TCP_CONG_NON_RESTRICTED 0x1 696 #define TCP_CONG_RTT_STAMP 0x2 697 698 struct tcp_congestion_ops { 699 struct list_head list; 700 unsigned long flags; 701 702 /* initialize private data (optional) */ 703 void (*init)(struct sock *sk); 704 /* cleanup private data (optional) */ 705 void (*release)(struct sock *sk); 706 707 /* return slow start threshold (required) */ 708 u32 (*ssthresh)(struct sock *sk); 709 /* lower bound for congestion window (optional) */ 710 u32 (*min_cwnd)(const struct sock *sk); 711 /* do new cwnd calculation (required) */ 712 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 713 /* call before changing ca_state (optional) */ 714 void (*set_state)(struct sock *sk, u8 new_state); 715 /* call when cwnd event occurs (optional) */ 716 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 717 /* new value of cwnd after loss (optional) */ 718 u32 (*undo_cwnd)(struct sock *sk); 719 /* hook for packet ack accounting (optional) */ 720 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 721 /* get info for inet_diag (optional) */ 722 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 723 724 char name[TCP_CA_NAME_MAX]; 725 struct module *owner; 726 }; 727 728 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 729 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 730 731 extern void tcp_init_congestion_control(struct sock *sk); 732 extern void tcp_cleanup_congestion_control(struct sock *sk); 733 extern int tcp_set_default_congestion_control(const char *name); 734 extern void tcp_get_default_congestion_control(char *name); 735 extern void tcp_get_available_congestion_control(char *buf, size_t len); 736 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 737 extern int tcp_set_allowed_congestion_control(char *allowed); 738 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 739 extern void tcp_slow_start(struct tcp_sock *tp); 740 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 741 742 extern struct tcp_congestion_ops tcp_init_congestion_ops; 743 extern u32 tcp_reno_ssthresh(struct sock *sk); 744 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 745 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 746 extern struct tcp_congestion_ops tcp_reno; 747 748 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 749 { 750 struct inet_connection_sock *icsk = inet_csk(sk); 751 752 if (icsk->icsk_ca_ops->set_state) 753 icsk->icsk_ca_ops->set_state(sk, ca_state); 754 icsk->icsk_ca_state = ca_state; 755 } 756 757 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 758 { 759 const struct inet_connection_sock *icsk = inet_csk(sk); 760 761 if (icsk->icsk_ca_ops->cwnd_event) 762 icsk->icsk_ca_ops->cwnd_event(sk, event); 763 } 764 765 /* These functions determine how the current flow behaves in respect of SACK 766 * handling. SACK is negotiated with the peer, and therefore it can vary 767 * between different flows. 768 * 769 * tcp_is_sack - SACK enabled 770 * tcp_is_reno - No SACK 771 * tcp_is_fack - FACK enabled, implies SACK enabled 772 */ 773 static inline int tcp_is_sack(const struct tcp_sock *tp) 774 { 775 return tp->rx_opt.sack_ok; 776 } 777 778 static inline int tcp_is_reno(const struct tcp_sock *tp) 779 { 780 return !tcp_is_sack(tp); 781 } 782 783 static inline int tcp_is_fack(const struct tcp_sock *tp) 784 { 785 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 786 } 787 788 static inline void tcp_enable_fack(struct tcp_sock *tp) 789 { 790 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 791 } 792 793 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 794 { 795 return tp->sacked_out + tp->lost_out; 796 } 797 798 /* This determines how many packets are "in the network" to the best 799 * of our knowledge. In many cases it is conservative, but where 800 * detailed information is available from the receiver (via SACK 801 * blocks etc.) we can make more aggressive calculations. 802 * 803 * Use this for decisions involving congestion control, use just 804 * tp->packets_out to determine if the send queue is empty or not. 805 * 806 * Read this equation as: 807 * 808 * "Packets sent once on transmission queue" MINUS 809 * "Packets left network, but not honestly ACKed yet" PLUS 810 * "Packets fast retransmitted" 811 */ 812 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 813 { 814 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 815 } 816 817 #define TCP_INFINITE_SSTHRESH 0x7fffffff 818 819 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 820 { 821 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 822 } 823 824 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 825 * The exception is rate halving phase, when cwnd is decreasing towards 826 * ssthresh. 827 */ 828 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 829 { 830 const struct tcp_sock *tp = tcp_sk(sk); 831 832 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 833 return tp->snd_ssthresh; 834 else 835 return max(tp->snd_ssthresh, 836 ((tp->snd_cwnd >> 1) + 837 (tp->snd_cwnd >> 2))); 838 } 839 840 /* Use define here intentionally to get WARN_ON location shown at the caller */ 841 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 842 843 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 844 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 845 846 /* The maximum number of MSS of available cwnd for which TSO defers 847 * sending if not using sysctl_tcp_tso_win_divisor. 848 */ 849 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 850 { 851 return 3; 852 } 853 854 /* Slow start with delack produces 3 packets of burst, so that 855 * it is safe "de facto". This will be the default - same as 856 * the default reordering threshold - but if reordering increases, 857 * we must be able to allow cwnd to burst at least this much in order 858 * to not pull it back when holes are filled. 859 */ 860 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 861 { 862 return tp->reordering; 863 } 864 865 /* Returns end sequence number of the receiver's advertised window */ 866 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 867 { 868 return tp->snd_una + tp->snd_wnd; 869 } 870 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 871 872 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 873 const struct sk_buff *skb) 874 { 875 if (skb->len < mss) 876 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 877 } 878 879 static inline void tcp_check_probe_timer(struct sock *sk) 880 { 881 const struct tcp_sock *tp = tcp_sk(sk); 882 const struct inet_connection_sock *icsk = inet_csk(sk); 883 884 if (!tp->packets_out && !icsk->icsk_pending) 885 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 886 icsk->icsk_rto, TCP_RTO_MAX); 887 } 888 889 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 890 { 891 tp->snd_wl1 = seq; 892 } 893 894 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 895 { 896 tp->snd_wl1 = seq; 897 } 898 899 /* 900 * Calculate(/check) TCP checksum 901 */ 902 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 903 __be32 daddr, __wsum base) 904 { 905 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 906 } 907 908 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 909 { 910 return __skb_checksum_complete(skb); 911 } 912 913 static inline int tcp_checksum_complete(struct sk_buff *skb) 914 { 915 return !skb_csum_unnecessary(skb) && 916 __tcp_checksum_complete(skb); 917 } 918 919 /* Prequeue for VJ style copy to user, combined with checksumming. */ 920 921 static inline void tcp_prequeue_init(struct tcp_sock *tp) 922 { 923 tp->ucopy.task = NULL; 924 tp->ucopy.len = 0; 925 tp->ucopy.memory = 0; 926 skb_queue_head_init(&tp->ucopy.prequeue); 927 #ifdef CONFIG_NET_DMA 928 tp->ucopy.dma_chan = NULL; 929 tp->ucopy.wakeup = 0; 930 tp->ucopy.pinned_list = NULL; 931 tp->ucopy.dma_cookie = 0; 932 #endif 933 } 934 935 /* Packet is added to VJ-style prequeue for processing in process 936 * context, if a reader task is waiting. Apparently, this exciting 937 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 938 * failed somewhere. Latency? Burstiness? Well, at least now we will 939 * see, why it failed. 8)8) --ANK 940 * 941 * NOTE: is this not too big to inline? 942 */ 943 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 944 { 945 struct tcp_sock *tp = tcp_sk(sk); 946 947 if (sysctl_tcp_low_latency || !tp->ucopy.task) 948 return 0; 949 950 __skb_queue_tail(&tp->ucopy.prequeue, skb); 951 tp->ucopy.memory += skb->truesize; 952 if (tp->ucopy.memory > sk->sk_rcvbuf) { 953 struct sk_buff *skb1; 954 955 BUG_ON(sock_owned_by_user(sk)); 956 957 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 958 sk_backlog_rcv(sk, skb1); 959 NET_INC_STATS_BH(sock_net(sk), 960 LINUX_MIB_TCPPREQUEUEDROPPED); 961 } 962 963 tp->ucopy.memory = 0; 964 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 965 wake_up_interruptible_sync_poll(sk_sleep(sk), 966 POLLIN | POLLRDNORM | POLLRDBAND); 967 if (!inet_csk_ack_scheduled(sk)) 968 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 969 (3 * tcp_rto_min(sk)) / 4, 970 TCP_RTO_MAX); 971 } 972 return 1; 973 } 974 975 976 #undef STATE_TRACE 977 978 #ifdef STATE_TRACE 979 static const char *statename[]={ 980 "Unused","Established","Syn Sent","Syn Recv", 981 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 982 "Close Wait","Last ACK","Listen","Closing" 983 }; 984 #endif 985 extern void tcp_set_state(struct sock *sk, int state); 986 987 extern void tcp_done(struct sock *sk); 988 989 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 990 { 991 rx_opt->dsack = 0; 992 rx_opt->num_sacks = 0; 993 } 994 995 /* Determine a window scaling and initial window to offer. */ 996 extern void tcp_select_initial_window(int __space, __u32 mss, 997 __u32 *rcv_wnd, __u32 *window_clamp, 998 int wscale_ok, __u8 *rcv_wscale, 999 __u32 init_rcv_wnd); 1000 1001 static inline int tcp_win_from_space(int space) 1002 { 1003 return sysctl_tcp_adv_win_scale<=0 ? 1004 (space>>(-sysctl_tcp_adv_win_scale)) : 1005 space - (space>>sysctl_tcp_adv_win_scale); 1006 } 1007 1008 /* Note: caller must be prepared to deal with negative returns */ 1009 static inline int tcp_space(const struct sock *sk) 1010 { 1011 return tcp_win_from_space(sk->sk_rcvbuf - 1012 atomic_read(&sk->sk_rmem_alloc)); 1013 } 1014 1015 static inline int tcp_full_space(const struct sock *sk) 1016 { 1017 return tcp_win_from_space(sk->sk_rcvbuf); 1018 } 1019 1020 static inline void tcp_openreq_init(struct request_sock *req, 1021 struct tcp_options_received *rx_opt, 1022 struct sk_buff *skb) 1023 { 1024 struct inet_request_sock *ireq = inet_rsk(req); 1025 1026 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1027 req->cookie_ts = 0; 1028 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1029 req->mss = rx_opt->mss_clamp; 1030 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1031 ireq->tstamp_ok = rx_opt->tstamp_ok; 1032 ireq->sack_ok = rx_opt->sack_ok; 1033 ireq->snd_wscale = rx_opt->snd_wscale; 1034 ireq->wscale_ok = rx_opt->wscale_ok; 1035 ireq->acked = 0; 1036 ireq->ecn_ok = 0; 1037 ireq->rmt_port = tcp_hdr(skb)->source; 1038 ireq->loc_port = tcp_hdr(skb)->dest; 1039 } 1040 1041 extern void tcp_enter_memory_pressure(struct sock *sk); 1042 1043 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1044 { 1045 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1046 } 1047 1048 static inline int keepalive_time_when(const struct tcp_sock *tp) 1049 { 1050 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1051 } 1052 1053 static inline int keepalive_probes(const struct tcp_sock *tp) 1054 { 1055 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1056 } 1057 1058 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1059 { 1060 const struct inet_connection_sock *icsk = &tp->inet_conn; 1061 1062 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1063 tcp_time_stamp - tp->rcv_tstamp); 1064 } 1065 1066 static inline int tcp_fin_time(const struct sock *sk) 1067 { 1068 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1069 const int rto = inet_csk(sk)->icsk_rto; 1070 1071 if (fin_timeout < (rto << 2) - (rto >> 1)) 1072 fin_timeout = (rto << 2) - (rto >> 1); 1073 1074 return fin_timeout; 1075 } 1076 1077 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1078 int paws_win) 1079 { 1080 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1081 return 1; 1082 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1083 return 1; 1084 /* 1085 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1086 * then following tcp messages have valid values. Ignore 0 value, 1087 * or else 'negative' tsval might forbid us to accept their packets. 1088 */ 1089 if (!rx_opt->ts_recent) 1090 return 1; 1091 return 0; 1092 } 1093 1094 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1095 int rst) 1096 { 1097 if (tcp_paws_check(rx_opt, 0)) 1098 return 0; 1099 1100 /* RST segments are not recommended to carry timestamp, 1101 and, if they do, it is recommended to ignore PAWS because 1102 "their cleanup function should take precedence over timestamps." 1103 Certainly, it is mistake. It is necessary to understand the reasons 1104 of this constraint to relax it: if peer reboots, clock may go 1105 out-of-sync and half-open connections will not be reset. 1106 Actually, the problem would be not existing if all 1107 the implementations followed draft about maintaining clock 1108 via reboots. Linux-2.2 DOES NOT! 1109 1110 However, we can relax time bounds for RST segments to MSL. 1111 */ 1112 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1113 return 0; 1114 return 1; 1115 } 1116 1117 static inline void tcp_mib_init(struct net *net) 1118 { 1119 /* See RFC 2012 */ 1120 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1121 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1122 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1123 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1124 } 1125 1126 /* from STCP */ 1127 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1128 { 1129 tp->lost_skb_hint = NULL; 1130 tp->scoreboard_skb_hint = NULL; 1131 } 1132 1133 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1134 { 1135 tcp_clear_retrans_hints_partial(tp); 1136 tp->retransmit_skb_hint = NULL; 1137 } 1138 1139 /* MD5 Signature */ 1140 struct crypto_hash; 1141 1142 union tcp_md5_addr { 1143 struct in_addr a4; 1144 #if IS_ENABLED(CONFIG_IPV6) 1145 struct in6_addr a6; 1146 #endif 1147 }; 1148 1149 /* - key database */ 1150 struct tcp_md5sig_key { 1151 struct hlist_node node; 1152 u8 keylen; 1153 u8 family; /* AF_INET or AF_INET6 */ 1154 union tcp_md5_addr addr; 1155 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1156 struct rcu_head rcu; 1157 }; 1158 1159 /* - sock block */ 1160 struct tcp_md5sig_info { 1161 struct hlist_head head; 1162 struct rcu_head rcu; 1163 }; 1164 1165 /* - pseudo header */ 1166 struct tcp4_pseudohdr { 1167 __be32 saddr; 1168 __be32 daddr; 1169 __u8 pad; 1170 __u8 protocol; 1171 __be16 len; 1172 }; 1173 1174 struct tcp6_pseudohdr { 1175 struct in6_addr saddr; 1176 struct in6_addr daddr; 1177 __be32 len; 1178 __be32 protocol; /* including padding */ 1179 }; 1180 1181 union tcp_md5sum_block { 1182 struct tcp4_pseudohdr ip4; 1183 #if IS_ENABLED(CONFIG_IPV6) 1184 struct tcp6_pseudohdr ip6; 1185 #endif 1186 }; 1187 1188 /* - pool: digest algorithm, hash description and scratch buffer */ 1189 struct tcp_md5sig_pool { 1190 struct hash_desc md5_desc; 1191 union tcp_md5sum_block md5_blk; 1192 }; 1193 1194 /* - functions */ 1195 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1196 const struct sock *sk, 1197 const struct request_sock *req, 1198 const struct sk_buff *skb); 1199 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1200 int family, const u8 *newkey, 1201 u8 newkeylen, gfp_t gfp); 1202 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1203 int family); 1204 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1205 struct sock *addr_sk); 1206 1207 #ifdef CONFIG_TCP_MD5SIG 1208 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1209 const union tcp_md5_addr *addr, int family); 1210 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1211 #else 1212 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1213 const union tcp_md5_addr *addr, 1214 int family) 1215 { 1216 return NULL; 1217 } 1218 #define tcp_twsk_md5_key(twsk) NULL 1219 #endif 1220 1221 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *); 1222 extern void tcp_free_md5sig_pool(void); 1223 1224 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1225 extern void tcp_put_md5sig_pool(void); 1226 1227 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1228 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1229 unsigned header_len); 1230 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1231 const struct tcp_md5sig_key *key); 1232 1233 /* write queue abstraction */ 1234 static inline void tcp_write_queue_purge(struct sock *sk) 1235 { 1236 struct sk_buff *skb; 1237 1238 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1239 sk_wmem_free_skb(sk, skb); 1240 sk_mem_reclaim(sk); 1241 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1242 } 1243 1244 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1245 { 1246 return skb_peek(&sk->sk_write_queue); 1247 } 1248 1249 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1250 { 1251 return skb_peek_tail(&sk->sk_write_queue); 1252 } 1253 1254 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1255 const struct sk_buff *skb) 1256 { 1257 return skb_queue_next(&sk->sk_write_queue, skb); 1258 } 1259 1260 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1261 const struct sk_buff *skb) 1262 { 1263 return skb_queue_prev(&sk->sk_write_queue, skb); 1264 } 1265 1266 #define tcp_for_write_queue(skb, sk) \ 1267 skb_queue_walk(&(sk)->sk_write_queue, skb) 1268 1269 #define tcp_for_write_queue_from(skb, sk) \ 1270 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1271 1272 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1273 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1274 1275 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1276 { 1277 return sk->sk_send_head; 1278 } 1279 1280 static inline bool tcp_skb_is_last(const struct sock *sk, 1281 const struct sk_buff *skb) 1282 { 1283 return skb_queue_is_last(&sk->sk_write_queue, skb); 1284 } 1285 1286 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1287 { 1288 if (tcp_skb_is_last(sk, skb)) 1289 sk->sk_send_head = NULL; 1290 else 1291 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1292 } 1293 1294 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1295 { 1296 if (sk->sk_send_head == skb_unlinked) 1297 sk->sk_send_head = NULL; 1298 } 1299 1300 static inline void tcp_init_send_head(struct sock *sk) 1301 { 1302 sk->sk_send_head = NULL; 1303 } 1304 1305 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1306 { 1307 __skb_queue_tail(&sk->sk_write_queue, skb); 1308 } 1309 1310 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1311 { 1312 __tcp_add_write_queue_tail(sk, skb); 1313 1314 /* Queue it, remembering where we must start sending. */ 1315 if (sk->sk_send_head == NULL) { 1316 sk->sk_send_head = skb; 1317 1318 if (tcp_sk(sk)->highest_sack == NULL) 1319 tcp_sk(sk)->highest_sack = skb; 1320 } 1321 } 1322 1323 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1324 { 1325 __skb_queue_head(&sk->sk_write_queue, skb); 1326 } 1327 1328 /* Insert buff after skb on the write queue of sk. */ 1329 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1330 struct sk_buff *buff, 1331 struct sock *sk) 1332 { 1333 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1334 } 1335 1336 /* Insert new before skb on the write queue of sk. */ 1337 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1338 struct sk_buff *skb, 1339 struct sock *sk) 1340 { 1341 __skb_queue_before(&sk->sk_write_queue, skb, new); 1342 1343 if (sk->sk_send_head == skb) 1344 sk->sk_send_head = new; 1345 } 1346 1347 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1348 { 1349 __skb_unlink(skb, &sk->sk_write_queue); 1350 } 1351 1352 static inline int tcp_write_queue_empty(struct sock *sk) 1353 { 1354 return skb_queue_empty(&sk->sk_write_queue); 1355 } 1356 1357 static inline void tcp_push_pending_frames(struct sock *sk) 1358 { 1359 if (tcp_send_head(sk)) { 1360 struct tcp_sock *tp = tcp_sk(sk); 1361 1362 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1363 } 1364 } 1365 1366 /* Start sequence of the skb just after the highest skb with SACKed 1367 * bit, valid only if sacked_out > 0 or when the caller has ensured 1368 * validity by itself. 1369 */ 1370 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1371 { 1372 if (!tp->sacked_out) 1373 return tp->snd_una; 1374 1375 if (tp->highest_sack == NULL) 1376 return tp->snd_nxt; 1377 1378 return TCP_SKB_CB(tp->highest_sack)->seq; 1379 } 1380 1381 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1382 { 1383 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1384 tcp_write_queue_next(sk, skb); 1385 } 1386 1387 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1388 { 1389 return tcp_sk(sk)->highest_sack; 1390 } 1391 1392 static inline void tcp_highest_sack_reset(struct sock *sk) 1393 { 1394 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1395 } 1396 1397 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1398 static inline void tcp_highest_sack_combine(struct sock *sk, 1399 struct sk_buff *old, 1400 struct sk_buff *new) 1401 { 1402 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1403 tcp_sk(sk)->highest_sack = new; 1404 } 1405 1406 /* Determines whether this is a thin stream (which may suffer from 1407 * increased latency). Used to trigger latency-reducing mechanisms. 1408 */ 1409 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1410 { 1411 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1412 } 1413 1414 /* /proc */ 1415 enum tcp_seq_states { 1416 TCP_SEQ_STATE_LISTENING, 1417 TCP_SEQ_STATE_OPENREQ, 1418 TCP_SEQ_STATE_ESTABLISHED, 1419 TCP_SEQ_STATE_TIME_WAIT, 1420 }; 1421 1422 int tcp_seq_open(struct inode *inode, struct file *file); 1423 1424 struct tcp_seq_afinfo { 1425 char *name; 1426 sa_family_t family; 1427 const struct file_operations *seq_fops; 1428 struct seq_operations seq_ops; 1429 }; 1430 1431 struct tcp_iter_state { 1432 struct seq_net_private p; 1433 sa_family_t family; 1434 enum tcp_seq_states state; 1435 struct sock *syn_wait_sk; 1436 int bucket, offset, sbucket, num, uid; 1437 loff_t last_pos; 1438 }; 1439 1440 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1441 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1442 1443 extern struct request_sock_ops tcp_request_sock_ops; 1444 extern struct request_sock_ops tcp6_request_sock_ops; 1445 1446 extern void tcp_v4_destroy_sock(struct sock *sk); 1447 1448 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1449 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, 1450 netdev_features_t features); 1451 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1452 struct sk_buff *skb); 1453 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1454 struct sk_buff *skb); 1455 extern int tcp_gro_complete(struct sk_buff *skb); 1456 extern int tcp4_gro_complete(struct sk_buff *skb); 1457 1458 #ifdef CONFIG_PROC_FS 1459 extern int tcp4_proc_init(void); 1460 extern void tcp4_proc_exit(void); 1461 #endif 1462 1463 /* TCP af-specific functions */ 1464 struct tcp_sock_af_ops { 1465 #ifdef CONFIG_TCP_MD5SIG 1466 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1467 struct sock *addr_sk); 1468 int (*calc_md5_hash) (char *location, 1469 struct tcp_md5sig_key *md5, 1470 const struct sock *sk, 1471 const struct request_sock *req, 1472 const struct sk_buff *skb); 1473 int (*md5_parse) (struct sock *sk, 1474 char __user *optval, 1475 int optlen); 1476 #endif 1477 }; 1478 1479 struct tcp_request_sock_ops { 1480 #ifdef CONFIG_TCP_MD5SIG 1481 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1482 struct request_sock *req); 1483 int (*calc_md5_hash) (char *location, 1484 struct tcp_md5sig_key *md5, 1485 const struct sock *sk, 1486 const struct request_sock *req, 1487 const struct sk_buff *skb); 1488 #endif 1489 }; 1490 1491 /* Using SHA1 for now, define some constants. 1492 */ 1493 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1494 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1495 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1496 1497 extern int tcp_cookie_generator(u32 *bakery); 1498 1499 /** 1500 * struct tcp_cookie_values - each socket needs extra space for the 1501 * cookies, together with (optional) space for any SYN data. 1502 * 1503 * A tcp_sock contains a pointer to the current value, and this is 1504 * cloned to the tcp_timewait_sock. 1505 * 1506 * @cookie_pair: variable data from the option exchange. 1507 * 1508 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1509 * indicates default (sysctl_tcp_cookie_size). 1510 * After cookie sent, remembers size of cookie. 1511 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1512 * 1513 * @s_data_desired: user specified tcpct_s_data_desired. When the 1514 * constant payload is specified (@s_data_constant), 1515 * holds its length instead. 1516 * Range 0 to TCP_MSS_DESIRED. 1517 * 1518 * @s_data_payload: constant data that is to be included in the 1519 * payload of SYN or SYNACK segments when the 1520 * cookie option is present. 1521 */ 1522 struct tcp_cookie_values { 1523 struct kref kref; 1524 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1525 u8 cookie_pair_size; 1526 u8 cookie_desired; 1527 u16 s_data_desired:11, 1528 s_data_constant:1, 1529 s_data_in:1, 1530 s_data_out:1, 1531 s_data_unused:2; 1532 u8 s_data_payload[0]; 1533 }; 1534 1535 static inline void tcp_cookie_values_release(struct kref *kref) 1536 { 1537 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1538 } 1539 1540 /* The length of constant payload data. Note that s_data_desired is 1541 * overloaded, depending on s_data_constant: either the length of constant 1542 * data (returned here) or the limit on variable data. 1543 */ 1544 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1545 { 1546 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1547 ? tp->cookie_values->s_data_desired 1548 : 0; 1549 } 1550 1551 /** 1552 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1553 * 1554 * As tcp_request_sock has already been extended in other places, the 1555 * only remaining method is to pass stack values along as function 1556 * parameters. These parameters are not needed after sending SYNACK. 1557 * 1558 * @cookie_bakery: cryptographic secret and message workspace. 1559 * 1560 * @cookie_plus: bytes in authenticator/cookie option, copied from 1561 * struct tcp_options_received (above). 1562 */ 1563 struct tcp_extend_values { 1564 struct request_values rv; 1565 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1566 u8 cookie_plus:6, 1567 cookie_out_never:1, 1568 cookie_in_always:1; 1569 }; 1570 1571 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1572 { 1573 return (struct tcp_extend_values *)rvp; 1574 } 1575 1576 extern void tcp_v4_init(void); 1577 extern void tcp_init(void); 1578 1579 #endif /* _TCP_H */ 1580